Bed load transport estimations in Goodwin creek using neural network methods

Abstract

Equations used for calculating bedload transport rates are generally developed based on the assumption of steady flow conditions.This implies that the relationship between bedload transport, discharge, flow depth, and shear stress is single-valued. One of thereasons for adopting such an approach is that almost all the pertinent laboratory data on bedload transport have been obtained fromexperiments performed under steady flow conditions. Similarly, the scarcity of accurate bed load field data obtained during thepassage of floods is attributed to the difficulties, which at times can become life-threatening, encountered under such conditions.Provision of data under challenging conditions may lead to the inability to provide data in some cases and interruption in datacontinuity. It is difficult to make predictions using classical statistical science in discontinuous or lack of data situations. Artificialneural networks (ANN) are useful in predicting when the data is insufficient. In this study, two frequently used ANN applications,radial basis functions, and generalized regression neural network are employed to estimate the bed load data. It was seen that theANN estimations are more satisfactory compared to those of the conventional statistical methods results. It was shown that ANNestimations for gravel bedload data are more successful than the sand load data.